Is AI our only hope for the future of humankind? Professor Sami Kaski explores how this powerful tool could help meet the challenges facing our world. But how do we ensure that the human is present in the machine?
Artificial Intelligence (AI) is all around us. From the smart watches we wear everyday collecting our personal biodata to helping medical professionals prescribe to patients, this technology has the potential to greatly advance global health services, to name just one area, in the future.
The University of Manchester is playing an important part in these advancements, as a world-leader in developing autonomous systems that combine AI technologies with robotics. An example is our work to improve the capabilities of robots used in the hazardous work to decommission legacy nuclear power reactors and helping to better protect surrounding communities.
Through machine learning, these robots obey their algorithms – the embedded digital instructions defining an AI task – and adapt performance using devolved decision-making capabilities, so they can be agile and respond to changing or complex environments.
This is an exciting period for research as we explore new frontiers of artificial intelligence
It’s impressive technology, but automated operation is largely limited by the need for human intervention, to set objectives and rewards within the algorithm that tell the system the desired outcomes. Why is this a challenge? Because as humans we often don’t have a fully defined goal at the beginning of a research project. Without a said goal, these intelligent machines can’t reach their full potential.
Pushing the frontier of AI design
Now it’s time for a bold vison. We need to take machines beyond simply responding to our literal inputs and develop systems that use intelligence to infer what’s really being asked of them by their imperfect human users.
“We’ve given algorithms a free pass for far too long,” says AI expert Professor Stuart Russell (from The promises and perils of AI broadcast by Radio Davos). We must push for full automation in our AI systems and put the human back into the algorithm.
But what about social responsibility, one of our University’s core values? If future AI is to act in a responsible way, they need to ‘think’: ‘what does this human user need from the task – they are not quite sure what they want, so how can I help?’
Formulating an intuitive model
At Manchester, we’re developing an intuitive AI that predicts the behaviour and expectations of its user, in line with the ‘Theory of Mind’. According to Theory of Mind, humans make assumptions about what others want, think and believe. We infer these states of mind, even though they’re not directly observable.
These inferences are the foundational elements for successful human social interaction; our AI will incorporate a similar approach to enable successful human-AI interactions.
Making an impact
This pioneering AI model is set to make a big impact across a number of industries. Human-inclusive machine learning will help us improve outcomes step-by-step and achieve trustworthy solutions to challenges. Say we wanted to design a personalised treatment for lung cancer, we would need to maximise the efficacy of radiotherapy while also minimising its side effects. To achieve this, doctors could use AI to combine their expertise with data learnings collected from earlier patient experiences.
Manchester has a long history of innovation in AI, stemming from Alan Turing’s time here in the 1940s. Today our researchers continue in the same spirit – both in research and the use of AI to solve problems across other fields. It’s even comparable to the start-up culture you might associate with new tech enterprises. With national and local government plans to ‘level up’ innovation and R&D investment across the UK, we could be witnessing the birth of a Silicon Valley in the north.
Inclusive by design
The Manchester AI team are not just putting the human back into the algorithm but, in fact, a diverse range of humans. Our researchers engage with a broad spectrum of end-users at the start of the design process – bringing greater representation and democracy to the development of human-inclusive AI systems. This is an exciting period for research as we explore new frontiers of artificial intelligence and start to better balance its relationship with humans.
Find out more about cutting-edge AI research at the Manchester Centre for AI Fundamentals, part of the Manchester Institute for Data Science and Artificial Intelligence.